Beamforming Design of IRS-Assisted SWIPT System Based on Deep Reinforcement Learning

In the intelligent reflecting surface (IRS)-assisted simultaneous wireless information and power transfer (SWIPT) system, it is critical to improve the system performance by jointly designing the transmit beamforming of base station and IRS. According to this beamforming design problem, the actual phase-related amplitude model is established that is affected by the nonlinear energy loss of IRS, and then a multiparameter joint optimization problem is established to maximize the rate sum of the devices. Furthermore, a PER-SAC algorithm is proposed to solve this problem, which is based on the soft actor–critic (SAC) principle and uses prioritized experience replay (PER) to improve training efficiency. The results show that the PER-SAC algorithm has better stability and convergence effect than other algorithms in this scenario, and the sum rate obtained by the PER-SAC algorithm is also higher than that of the random phase-shift scheme and the random power splitting ratio scheme.